Andrew Beekhof a écrit :
What it wont do is move to a node where the resource has _already_
failed.
That's my point.
The rationale is that if something happened on nodeX enough times to
cause the resource migrate away, then there is unlikely to be any
point trying nodeX again until the admin has fixed the underlying
problem (and re-set the failcount).
I understand. But in my point of view, I think it will be more
interesting to know if the resource can be run again on the primary node
than to know that it has already failed X times.
In that way, the "automagically reset of the failcount" (even if it is
not the right thing) using a timer does not answer at the question "Is
the resource may run again" (e.g. the IP connexion is back for example).
We want to avoid situations where resources bounce around the cluster
(causing undue churn and load) indefinitely without any chance of
actually running anywhere.
It sounds good and a timer seems to be a right way.
Note also that this is all talking about resource-level failures and
does not apply if the whole node fails.
Sure.
Thank for all.
--
Hervé GAUTIER
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
